frizynn/qwen3-think-235B-A22B-Thinking-2507-2bit-UD-Q2_K_XL

⛓️ Library: transformers
💾 Memory requirement: ~86.5 GiB to load and run in Ollama
🛠️ Tags: qwen, qwen3, unsloth

Highlights

Over the past quarter, Unsloth’s Qwen3-235B-A22B-Thinking-2507 has seen major improvements in deep reasoning, instruction following, and long‑context capabilities:

• Thinking mode tuned for logic, math, coding and scientific benchmarks
  
• State‑of‑the‑art open‑source results on chain‑of‑thought tasks
  
• 256 K context window native support for very long documents
  

Model Overview

Quickstart: Ollama

# Install Ollama (Linux)
curl -fsSL https://ollama.com/install.sh | sh

# Pull the model into Ollama
ollama pull frizynn/qwen3-think-235B-A22B-Thinking-2507-2bit-UD-Q2_K_XL

# Run an interactive Ollama session
ollama run frizynn/qwen3-think-235B-A22B-Thinking-2507-2bit-UD-Q2_K_XL

# ⚠️ Requires ~86.5 GiB system RAM

Example: LangChain v0.3 integration

from langchain_core.prompts import ChatPromptTemplate
from langchain_ollama.llms import OllamaLLM

# Define a simple step-by-step prompt
template = """User: {question}

Assistant (thinking):"""

prompt = ChatPromptTemplate.from_template(template)

# Point to your local Ollama model
llm = OllamaLLM(model="frizynn/qwen3-think-235B-A22B-Thinking-2507-2bit-UD-Q2_K_XL")

# Chain prompt and LLM
chain = prompt | llm

# Invoke with a question
response = chain.invoke({"question": "What are the benefits of chain‑of‑thought prompting?"})
print(response)

Memory & Performance Snapshot

Credits:

• Model by Qwen Team (Alibaba)
• Quantization via Unsloth Dynamic 2.0
• Runtime powered by Ollama
• Example integration with LangChain v0.3